low latency requirements for ng-epongrouper.ieee.org/groups/802/3/ca/public/meeting...5 march 2017...

Jun Shan Wey

Li Quan Yuan

Wei Liang Zhang

Low Latency Services and Requirements for 100G EPONIEEE P802.3ca Task Force Meeting, March 2017

Vancouver BC, Canada

March 2017 IEEE P802.3ca Task Force Meeting, Vancouver BC, Canada

In the Huntington Beach meeting, we discussed future services, which

might require low latency and impact how standards should be

specified (wey_3ca_01_0117)

Based on feedback from members, there is interest in further

investigation of latency related topics

This contribution provides more detail on both bandwidth and latency

requirements of mobile fronthaul for different 5G services and of

virtual reality/augmented reality video streaming

We identified topics to develop in standards to support these services

Motivation and purpose of this contribution


Mobile Fronthaul Evolution

• What is mobile fronthaul?• How much bandwidth do we need?• What is the latency requirement?• What is the recommended path forward?

4 March 2017 IEEE P802.3ca Task Force Meeting, Vancouver BC, Canada

Traditional fronthaul link in Radio Access Network

RU: remote Radio UnitDU: baseband Digital UnitS-GW: service gatewayMME: mobile management entity

Fronthaul CPRI/OBSAI (fiber, copper)

< 100 meters

Central Office

Cell site

DU

Traditional Backhaul (copper, microwave, PtP fiber)

Antenna

Backhaul aggregation

S-GW/MME

RU

• CPRI Fronthaul line rate depends on many factors, e.g., number of antennas and sectors, sampling rate, line coding, etc. CPRI Option 10 specifies 24.33Gbps for a 20MHz signal. For a 100 MHz signal, 3 sectors, and 8 antenna/sector, the line rate could be 148 Gbps!

• As values of all the related factors are expected increase drastically in 5G New Radio, it will be extremely difficult to support the CPRI fronthaul bandwidth using current PON systems

• NGMN Alliance recommended the total round-trip latency budget between cell site and the core network must be <10ms, and preferably <5ms. The delay budget allocated to the backhaul link is typically 1/3 of this budget

• Small Cell Forum classifies backhaul system latency as <1ms (good); 1-5ms (OK); >5ms (poor)


Mobile Fronthaul evolution towards Centralized/Cloud RAN

Central Office

Cell site

Metro aggregation

Fronthaul – new RANSplit Processing

Ethernet over fiber

DUpool Gateway

Backhaul

SCENARIO 1

SCENARIO 2

Central Office

Cell site

Metro aggregation

DU

DU

DU

BBU Cluster

Fronthaul – legacy CPRI/OBSAI over fiber

Cell site

Cell site

Backhaul


Capacity and latency requirements for Scenario 1

• Same capacity requirements as in the traditional case• Total round-trip delay = processing time in RU + 2x transit time in fiber + processing time in DU

- Max round-trip processing delay per link is 5 ms (CPRI spec v7.0, clause 7.1.8.1)- Max total round-trip delay between RU and DU is therefor ~105 ms/10km or ~210 ms/20km (note: round trip

delay in fiber is 10 ms/km)

• NGFI (next gen fronthaul interface) specification: - Transport equipment one-way delay is ~220 ms, which requires <10 ms one-way forwarding time per equipment

for a 20km link

Central Office

Cell site

Fronthaul – legacy CPRI/OBSAI over fiber

Metro aggregation

DU

DU

DU

DU Cluster

Cell site

Backhaul


Capacity and latency requirements for Scenario 2

Many potential functional split options!

RU/GWDU

Central Office

Cell site

Metro aggregation

DUpool Gateway

Backhaul

Cell site

Source: FSAN

Fronthaul – new RANSplit Processing

Ethernet over fiber


Capacity requirements for different functional split options

Source: FSAN

Functional Split Option

System Capacity for Different Signal Bandwidth

10 MHz 20 MHz 200 MHz 1GHz

Option 1 0.38 Gbps 0.76 Gbps 7.6 Gbps 38 Gbps




Option 5 0.4 Gbps 0.8 Gbps 8 Gbps 40 Gbps

1 2 3 4 5


NGMN 5G system latency requirements• NGMN stated the E2E RTP latency for a 5G

system could be < 1 ms. What are these use cases? Do they need to, can they, be supported by new generation PON?

• Ultra-low latency use case:• Tactile internet where humans will wirelessly

control real and virtual objects, manufacturing, remote medical care, autonomous cars

• Ultra-high reliability & ultra-low latency use case:• Collaborative robots in manufacturing: not valid• Automated traffic control and driving, remote

object manipulation (e.g. remote surgery)

• To support these machine type communications use cases, our estimate for the PON segment is 10-20 ms for round-trip latency not including the fiber path delay


Conclusion for 5G MFH

• Both capacity and latency requirements depend on the choice of functional split point• 100G-EPON should be able to support the MFH bandwidth requirements for the new

RAN scenario with split processing• Latency requirements of machine-type communications are extremely stringent:

estimate for the PON segment is 10-20 ms (round-trip delay) not including fiber path delay. New innovations will be needed

• Impact on specifications of channel bonding, downstream traffic scheduling, and DBA optimization should be considered

• IEEE 802.3ca should coordinate the effort with other SDOs to choose the preferred functional split option


Big Video Services

• How much bandwidth do we need to stream a VR video?• What is the latency requirement?• Can the existing network support a good VR experience?


How much bandwidth do we need to stream a VR video?

• Non-VR video stream with H.265 encoding (more detail in the appendix):• 4K format: 12-15 Mbps/video stream (OTT), 22.5-75 Mbps (IPTV)• 8K format: 48-60 Mbps/video stream (OTT), 90-300 Mbps (IPTV)

• VR video stream:• 4K format is the bare minimum starting point. 8K is preferred• Typical video format for VR is 2:1 as opposed to 16:9. The same video for regular TV is converted to 2:1 by

the camera or headset for VR viewing• Need two streams for stereoscopic experience: >600 Mbps/VR stream (1200 Mbps for VR+) could be needed• Other video encoding techniques to reduce file size are being explored, e.g., Cube Maps by Facebook


Can the existing network support a good VR experience?

• Existing network should be sufficient to support the latency requirement of VR video streaming

• Packet loss rate (1 error/8 hrs) is within expectation (<1.0 x 10-5) when tested in a G-PON network

• Interactive VR will have more stringent requirements, which is unknown at the moment. Synchronization between video and audio could add another dimension of complexity

OLTBRAS

CR

Level 3 CDN

Level 2CDN

IPBackbone

ONUSTB AGG SW

HOME ACCESS METRO BACKBONE

< 1ms < 1ms< 1ms

Future mini CDN

Format Bandwidth RTT Time Delay

Basic 4K 45Mbps < 20 ms

Basic 8K 180 Mbps < 16 ms

VR+ 1200 Mbps < 12 ms


Conclusion and proposal

• Mobile fronthaul/backhaul services for future 5G networks demand high capacity and low latency

• Big video services will require high capacity network. Interactive VR services have unknown stringent latency requirements

• Proposal of topics to further develop in standards:

₋ Further latency reduction in the case of channel bonding

₋ Optimize downstream traffic scheduling to reduce latency

₋ Optimize DBA to minimize latency: grants always ready for upstream traffic. Grant to one ONU could be limited to microsecond level

Thank You 谢谢！


Bandwidth and other requirements for different video formatsIPTV broadcast

Quasi 4K Basic 4K Ultra 4K Quasi 8K Basic 8K Ultra 8K VR VR+

Resolution3840x2160

3840x2160

3840x2160

7680x4320

7680x4320

7680x4320

3840x2160

7680x4320

Frame rate 30P 60P 120P 30P 60P 120P 120P 120P

Color depth 8bit 10bit 12bit 8bit 10bit 12bit 12bit 12bit

Compression algorithm

H.265 H.265 H.265 H.265 H.265 H.265 H.265 H.265

Average bitrate (bps)

15M 30M 50M 60M 120M 200M 200M 800M

Bandwidthrequirement

(bps)22.5M 45M 75M 90M 180M 300M 300M 1200M

low latency requirements for ng-epongrouper.ieee.org/groups/802/3/ca/public/meeting...5 march 2017...

Documents